{"about_md":"## Table of Contents\n- [Features](#features)\n- [Terms of use](#terms-of-use)\n- [Requirements](#requirements)\n- [Outputs](#outputs)\n- [Contributing](#contributing)\n- [License](#license)\n- [Contact](#contact)\n\n**Note**: predicTCR.com is currently in beta and serving predicTCR v1 model.\n\n## Features\npredicTCR v2 is a machine learning classifier pre-trained to identify tumor-reactive T cells from single cell sequencing (scRNA-seq) datasets. By including matched single cell TCR sequencing \ndata (scVDJ), predicTCR v2 can rapidly identify suitable TCRs for personalised cell therapies.  \n- **Accuracy:** predicTCR v2 outperforms other computational methods for determining tumor-reactive T cells,\n- **Tumor type agnostic:** predicTCR v2 works in diverse tumor types - including blood tumors,\n- **CD4 and CD8:** predicTCR v2 identifies tumor-reactive CD8 and CD4 T cells.\n\n\n## Terms of use\nThis service is provided for non-commercial use only, and is not to be used for training derivative models. By using the predicTCR v2 you agree to allow us to store the uploaded data to perform \nmeta-analyses of submitted data in order to continually improve the predicTCR v2 service. For non-commercial use please register an \naccount using your **academic email address**.  \n\nDue to the compute costs for each analysis, users are limited to the analysis of 10 samples. If you require more samples, or want to perform **commercial use**, please \n[contact us](mailto:predictcr@dkfz.de?subject=predicTCR%20extended%20access%20/%20commercial%20use%20request).\n\nIf you use predicTCR v2 please cite: 'Prediction of tumor-reactive T cell receptors from scRNA-seq data for personalized T cell therapy',\nNature Biotechnology 2024, DOI [10.1038/s41587-024-02161-y](https://www.nature.com/articles/s41587-024-02161-y). This helps us maintain the funding needed to keep the server running.\n\n\n## Requirements\n**Warning:**  \npredicTCR v2 does **not** perform any QC on your data. The principle of 'rubbish in = rubbish out' applies; please ensure that you upload matching scRNA-seq and scVDJ datasets. **Dataset with non-T cells will skew the automated thresholding.**\n\n**Input files:**\n- `Sample name`: each sample must be named - we recommend a unique name to help you track your data\n- `Tumor type`: please enter to tumor type or 'undisclosed'.  \nWe use this information to see whether we need to improve performance for particular tumor types.\n- `H5`: the data output from your scRNA-seq platform (**max 50mb**). \n    If you have non-h5 file (such as `.mtx`, you can save them to h5 file using `write10xCounts` from `DropletUtils` in `R`.\n- `CSV`: a CSV file containing information on TCR clonotypes, as in the 10x Genomics' Cell Ranger V(D)J seq outputs (**max 10mb**). Must contain at least three named columns:  \n    1. `barcode`: the barcode associated with each cell,  \n    2. `cdr3`: the cdr3 amino acid sequence associated with this cell,  \n    3. `chain`: whether this line specifies a TRA or TRB sequence.  \n\n**Data quality:**\nThe quality of the sequencing data is critical for generating high confidence predictions. Common problems include:\n- Tissue samples processed a long time after surgery, leading to degradation and loss of low abundance RNA transcripts >> *optimise tissue processing*,\n- Loss of transcripts due to poor sample handling during the scRNA-seq library preparation steps >> *switching to a different kit chemistry with better mRNA capture*,\n- Insufficient library sequencing depth  >> *repeat sequencing of remaining library material*,\n- Low numbers of T cells in sample >> *perform FACS prior to scSEQ capture*.\n\n**Example data:**\nFor testing we recommend using the processed and de-identified single-cell data from [Caushi et al 2021](https://pubmed.ncbi.nlm.nih.gov/34290408), \navailable in the Gene Expression Omnibus with accession number [GSE176022](http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE176022). \nFor comparison with published data use samples MD01-004, MD01-005, MD043-011.\n\n**Limits:**\nBy default the predicTCR v2 server will timeout if a samples takes longer than 10 minutes to analyse. \n\n## Outputs\n**Output modes:**\n 1. `Limited` (**default**): this reports the 10 most frequent TCR clonotypes which are called as reactive.   \n 2. `Detailed`: the reactivity score for each cell, please [contact us](mailto:predictcr@dkfz.de?subject=predicTCR%20detailed%20results%20request). to enable this (which may require an MTA agreement with the DKFZ).  \n\n**Output example:**  \n*coming soon.*  \n\n**Output errors:**  \npredicTCR v2 is written to fail gracefully and return informative error messages. If you are still stuck please [report an issue ](https://github.com/ssciwr/predicTCR/issues/new/choose).  \n\nIf you encounter a **\"permission denied\"** error when submitting a job, this might be due to a temporary internet connectivity lapse on our server.\n\n**Solution:** Please wait a few minutes and try submitting your job again. This issue is typically transient and resolves itself.\n\n## Questions\n1. Why is there a discrepancy between the number of cells per TCR compare to vloupe browser?\n- vloupe browser remove cells that are not considered cells / productive. You can remove those if desired before uploading your filtered contigs file.\n- vloupe does not match the cells to the gene expression whereas predicTCR requires both gene expression and TCR data. So predicTCR will exclude cells that does not contain both gene expression and TCR data.\n\n## Contributing\nThe predicTCR v2 web server is based on open-source code from the [Scientific Software Center](https://www.ssc.uni-heidelberg.de/en) of Heidelberg University.\nWe welcome contributions as pull requests or our [Github repository](https://github.com/ssciwr/predicTCR/). \n\n## License\n\nThe code underlying the predicTCR web service is Copyright (c) 2024 Liam Keegan and licensed under the [MIT License](https://github.com/ssciwr/predicTCR/blob/main/LICENSE).  \nModel weights for predicTCR are proprietary IP owned by the [German Cancer Research Centre (DKFZ)](https://www.dkfz.de/en/index.html).\n\n\n## Contact\n\n**Project Maintainer:**\n- Name: Chin Leng Tan\n- Email: [predictcr[at]dkfz.de](mailto:predictcr@dkfz.de)\n\nFeel free to reach out for any questions or collaboration opportunities in the areas of TCR discovery, TCR manufacture and experimental TCR validation.\n\n","csv_required_columns":"barcode;cdr3;chain","default_personal_submission_interval_mins":1,"default_personal_submission_quota":40,"global_quota":1241,"id":1,"max_filesize_csv_mb":20,"max_filesize_h5_mb":90,"news_items_json":"[{\"id\":\"1\",\"url\":\"\",\"text\":\"Note: The server is currently on maintenance.\"},{\"id\":\"2\",\"url\":\"https://www.nature.com/articles/s41467-025-61822-x\",\"text\":\"New paper in Nature Communication: click to read how predicTCR found tumor-reactive T cells in CLL patients!\"},{\"id\":\"3\",\"url\":\"https://maketcr.com\",\"text\":\"Stuck cloning TCRs? Click to try our our new tool 'makeTCR'!\"},{\"id\":\"4\",\"url\":\"https://www.nature.com/articles/s41591-024-03152-x\",\"text\":\"New paper in Nature Medicine: click to read how predicTCR found tumor-reactive T cells in bone marrow!\"},{\"id\":\"5\",\"url\":\"\",\"text\":\"Remember: predicTCR v2 is for research use only and has not been clinically validated\"},{\"id\":\"6\",\"url\":\"\",\"text\":\"predicTCR emails should now be delivered; let us know if you haven't yet received access\"}]","platforms":"10x_NextGEM;10x_GEM-X;BD_Rhapsody;ParseBio;PIPseq","runner_job_timeout_mins":30,"sources":"TIL (supported);Bone Marrow (supported); PMBC (in development)","tumor_types":"Undisclosed;Brain;Skin;Breast;Lung;Prostate;Bladder;Colorectal;PDAC;Blood;Ovarian"}